PM's Circumflex, Syntax and Philosophy of Types

9 PM’s Circumflex, Syntax and Philosophy of Types Kevin C. Klement From The Palgrave Centenary Companion to Principia Mathematica, edited by Nicholas Griffin and Bernard Linsky. New York: Palgrave Macmillan, 2013, pp. 218–246. (Author’s typesetting with matching pagination.) Introduction While second-order logic has its share of proponents, and specialized forms of type theory play a role in contemporary computer science and linguistics, I think it is fair to say that there’s relatively little contemporary interest in the sort of full-blown higher-order logic ex- emplified by the simple and ramified theories of types, at least for its own sake. One does not often see, for example, a new theory or devel- opment using it as its base system. The reasons for this are no doubt many. I think one major contributing factor, however, is a disconnect between the logicians who first advocated such an approach to logic and those who have been responsible for formulating it with modern standards of rigor. Principia Mathematica (PM), remains, it is fair to say, the best known exemplar of a type-theoretic approach to logic, but exactly what its type-theory is is far from agreed-upon. Whitehead and Russell are accused of unclarity, sloppiness or even outright con- fusion with regard to the syntax of their language, their system’s ax- iomatic foundations, and even its philosophical justification. More re- cent formulations of simple and ramified type-theories, such as those in Alonzo Church’s work, although formally unambiguous and irre- proachable, are seen as idiosyncratic and needlessly restrictive ways of codifying the “iterative conception” of sets or classes, more of a curiosity than a genuine rival to more flexible rival ways of codifying the same conception, such as ZFC and related set theories. But this is not surprising. The modern rigorous formulations have been done in detachment from, if not complete ignorance of, the real—or Russel- lian, at least—philosophical motivation of type theory, and are often done in ways that obscure that motivation. PM‘s Circumflex, Syntax and Philosophy of Types 219 It is not uncommon, for example, to find Church’s system of ramified types (or r-types), or something very similar, offered in place of an explanation of PM’s syntax (see, e.g., Church 1976, Linsky 1999, Urquhart 2003). Even if, contrary to what I shall argue, that system were equivalent to what PM was meant to be, or would become if properly reconstructed, to offer only that is unhelpful to would-be readers of PM. The syntax of that system is flatly unrecognizable in what one finds in PM itself. What is needed is an historically minded intepretation of the actual PM, but one that does not sacrifice contemporary standards of rigor. Ideally, this would consist first in a formulation of the syntax of PM, which, once the definitions, abbreviations and conventions adopted by Whitehead and Russell (Peano’s dot notation, the “typical ambiguity” method of suppressing type-indices, and so on) were accounted for, predicts precisely why and how the actual numbered propositions of PM appear the way they do. This should be presented alongside a philosophical explanation of the motivation for the type hierarchy. If successful, this project might show that even if Whitehead and Russell did not think of the formulation of a logical system and its syntax and semantics exactly the way contemporary logicians tend to, their approach had its own rhyme and reason. Contemporary prac- tices have been shaped to a large extent by the demands of logical meta-theory. While I do not agree with those who argue that Rus- sell’s views of logic are antithetical somehow to the very project of logical meta-theory,1 it was not his own focus. One cannot get very far in a metatheoretic proof without a full recursive definition of a well-formed expression of the object language, for example. If one’s aim is rather to use a given object language to state and demonstrate mathematical theorems, it perhaps suffices to make the notation clear enough that the mathematical content of those proofs is not obscured, leaving enough flexibility for refinements to the syntax to be made on the fly. Indeed, I suspect that Whitehead and Russell were deliber- ately less than fully explicit about the details of their system in hopes that the core of their mathematical proofs could be maintained even through substitution of a different precise understanding of the type system (cf. PM p. vii). Nevertheless, for the most part, it is still possi- ble to determine what they had in mind. This project, pursued in its entirety, is a large one. My aim here is a relatively modest one of getting clear about the syntax PM uses for expressions (variables and perhaps other terms) of higher-order: so-called “propositional functions”. My particular emphasis is on the use, or lack thereof, of the circumflex notation for function- abstraction. I argue that in many ways the notation used in PM is a 220 Kevin C. Klement kind of intermediate between the approach to the syntax of a theory of types found in Frege’s theory of functions of different levels, involving “incomplete” expressions with different structured kinds of incompleteness, and later, more familiar, devices for function abstraction, such as the λ-abstracts of the typed λ-calculus, where one can form complete terms for functions of any type. The discussion is one small part of trying to get a better handle on the philosophical justification, or perhaps inevitability, Russell thought there was for type-theory, a discussion which has been helped immensely in re- cent years by the availability of the surviving pre-PM manuscripts. I shall take for granted the conclusion reached not only by myself (Klement 2004, Klement 2010), but also by others (e.g., Landini 1998 and Stevens 2005), that these manuscripts show that even in 1910, Russell did not understand the type-hierarchy of PM as a hierarchy of entities of different logical kinds, whether those entities are to be understood as sets or classes at various stages of the iterative hierarchy or as abstract attributes-in-intension, “propositional functions” understood as mind- and language-independent real things. Indeed, my aim is largely to attempt to explain how this reading is compat- ible with taking the syntax of PM to include a very limited role for terms apparently standing “for” propositional functions, formed with the circumflex. In the appendices, I briefly sketch my reconstruction of the syntax and semantics of the 1910 first edition of PM (—I here bracket the question as to whether and to what extent the second edition is different—) though there are aspects to my reconstruction that re- quire more commentary and justification than what’s given here, par- ticularly with regard to those aspects that go beyond the discussion of the circumflex and notation for types. It is worth beginning our discussion with brief recaps of the con- trasting approaches. Frege’s approach On the Fregean approach, variables of different types (“levels” in Frege’s vocabulary) are literally of different shapes. A term for an object or an individual is a complete syntactic unit, whether simple or complex. An expression for a first-level function can be regarded as what is obtained from a complex term for an object by removing (one or more occurrences of) a simpler part which is itself a name for an object. In this way function expressions are gappy, incomplete, or as Frege says, “unsaturated”. PM‘s Circumflex, Syntax and Philosophy of Types 221 A second-level function expression is obtained by removing a first- level function expression from a complex term, and thus, while gappy, is not gappy in precisely the same way that a first-level function expression is. Only another incomplete expression can complete it. Compare the first-level function expression “F( )” and the second- level function expression “(9x) ::: x ::: ”.2 The latter’s argument expression, unlike the former’s, must itself have a place to receive the “x”. The two expressions mutually saturate. This difference is re- flected even when variables are used in Frege’s notation. For a variable for second-level concepts, Frege writes not (e.g.) “M”, but “Mβ : : : β : : : ” (Frege 1964, §25). Thus the canonical notation for a type-2 function taking as argument a variable type-1 function is: (1) Mβ f(β) If “f( )” here is instantiated to some complexly defined instance, such as “(F() _ G( ))”, the mutually saturating mark, “β” is placed inside both gaps to create: (2) Mβ(F(β) _ G(β)) If “Mβ : : : β : : : ” is itself instantiated to a complexly defined instance, such as “(x)(::: x ::: ⊃ ::: x ::: )”, the mutual saturation takes place over the whole: (3) (x)((F(x) _ G(x)) ⊃ (F(x) _ G(x))) The expression (3) is of the form “Mβ f(β)”, but multiply so, since it can also be seen as the result of giving f( ) and Mβ : : : β : : : the values ((F() _ G( )) ⊃ (F() _ G( ))) and (x) ::: x ::: , or the values G( ) and (x)((F(x) _ ::: x ::: ) ⊃ (F(x) _ ::: x ::: )), respectively, instead, and so on. This approach suggests a certain kind of philosophy for the levels hierarchy itself. There is not necessarily one privileged decom- position of an expression into function and argument. A function expression is not necessarily one unified “piece” of notation; there is no such thing as a “function term”, and hence, no such thing as placing a term of the wrong sort where a term of another sort ought to go.

Load more